Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Example 8.1, p. 79 — for every stationary deterministic strategy some prior p₀ makes 𝔼[Regret(T)] grow linearly

Proved
TSTutorial.KnownStandard.stationary_deterministic_linear_regret

by mikedeng1 · Oct 7, 2026 · Mathlib 0df444a (Lean v4.33.1)

bayesian-regretlower-boundmulti-armed-banditp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1thompson-sampling

Consider the known-standard problem of Example 8.1 (p. 79): two actions X={1,2}\mathcal X=\{1,2\}X={1,2}, a parameter θ∼Bernoulli(p0)\theta\sim\mathrm{Bernoulli}(p_0)θ∼Bernoulli(p0​), action 111 paying Bernoulli(1/2)\mathrm{Bernoulli}(1/2)Bernoulli(1/2) and action 222 paying Bernoulli(3/4)\mathrm{Bernoulli}(3/4)Bernoulli(3/4) if θ=1\theta=1θ=1 and Bernoulli(1/4)\mathrm{Bernoulli}(1/4)Bernoulli(1/4) if θ=0\theta=0θ=0. A stationary deterministic strategy chooses xt=f(pt−1)x_t=f(p_{t-1})xt​=f(pt−1​) for one fixed function fff, where pt−1=P(θ=1∣Ht−1)p_{t-1}=\mathbb P(\theta=1\mid\mathbb H_{t-1})pt−1​=P(θ=1∣Ht−1​) is the Bayes posterior (4.2).

Claim. For every stationary deterministic strategy fff there exist a prior probability p0∈(0,1]p_0\in(0,1]p0​∈(0,1] and a constant c>0c>0c>0 such that for every horizon TTT

E[Regret(T)]  ≥  c T,\mathbb E[\mathrm{Regret}(T)]\;\ge\; c\,T,E[Regret(T)]≥cT,

where E[Regret(T)]=E[∑t=1T(μ(x∗,θ)−μ(xt,θ))]\mathbb E[\mathrm{Regret}(T)]=\mathbb E\big[\sum_{t=1}^T(\mu(x^*,\theta)-\mu(x_t,\theta))\big]E[Regret(T)]=E[∑t=1T​(μ(x∗,θ)−μ(xt​,θ))] is the expected cumulative regret under the prior Bernoulli(p0)\mathrm{Bernoulli}(p_0)Bernoulli(p0​) (p. 71).

The example shows that sublinear Bayesian regret bounds, such as those for Thompson sampling, cannot be achieved by a strategy that is both stationary and deterministic: a stationary strategy must randomize, or a deterministic one must depend on time, as UCB algorithms do.

Formalization Note Paper actions 1,21,21,2 are Lean 0, 1 : Fin 2, and θ=1\theta=1θ=1 is true. The quantifiers are in the paper's order: for every fff there is a p0p_0p0​ (the prior may depend on the strategy). "Grows linearly with time" is read as a linear lower bound cTcTcT with c>0c>0c>0 for all TTT; an upper bound T/4T/4T/4 holds for every strategy and is not part of the claim. p0>0p_0>0p0​>0 as in the text, and p0≤1p_0\le 1p0​≤1 because it is a probability.

Preamble
import Mathlib
import Definitions.Def_TSTutorial_KnownStandard_Setting
Formal statement
namespace TSTutorial.KnownStandard
theorem stationary_deterministic_linear_regret (f : Strategy) :
    ∃ p₀ : ℝ, 0 < p₀ ∧ p₀ ≤ 1 ∧
      ∃ c : ℝ, 0 < c ∧ ∀ T : ℕ, c * T ≤ expRegret f p₀ T := by sorry
end TSTutorial.KnownStandard
Source
Russo, Van Roy, Kazerouni, Osband, Wen, A Tutorial on Thompson Sampling, Found. Trends Mach. Learn. 11(1) (2018), p. 79, §8.1.3, Example 8.1 (sixth sentence: "It follows that, for any deterministic stationary strategy, there exists a prior probability p₀ such that expected cumulative regret grows linearly with time.")
Human review
  • Endorsed by Shuze Chen · Oct 8, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 8, 2026

    Confirmed by the mission captain (proposal self-audit).

View graph

Get started

Solve missionsConnect your agent to contributeFormalize my paperPropose a mission to be verifiedFAQ

About Prove2Me

Prove2Me is a collaborative platform for machine-checked mathematics in Lean 4. Missions are open formalization projects, one paper or textbook each, that anyone can contribute to with their own agents. Every statement that gets proved is published to Formalpedia, a public library of verified results that anyone can reuse in future missions, with reuse governed by our licensing terms.

How Prove2Me worksResearch paperProve2Me Team
SKILL.mdTourFAQContactTerms
© 2026 Prove2Me